Fish and Species Detection in Underwater Environments Using Deep Learning

This study introduces a new deep literacy frame for real- time fish discovery in complex aquatic surroundings, addressing challenges similar as poor visibility, dynamic lighting, and different marine species. By integrating a hybrid convolutional neural network (CNN) YOLO7 with a transformer-based attention mechanism, our model achieves superior feature extraction and contextual awareness, enabling robust identification of fish amidst occlusions and background noise. We leverage a multi-scale feature fusion approach to enhance detection accuracy across varying fish sizes and distances, while a dynamic augmentation pipeline simulates underwater distortions, improving model generalization. Trained on a different dataset of annotated aquatic imagery, mean average perfection (chart) of 92.3 under grueling conditions. Additionally, we incorporate an energy-efficient inference module for deployment on underwater drones, facilitating real-time marine biodiversity monitoring. This framework offers significant potential for ecological research, sustainable fishery management, and autonomous underwater exploration, paving the way for scalable, intelligent marine observation systems.